First HARK model opens with jargon

[sbenthall]

I understand that:

• HARK is currently mainly used to teach a specific field of consumption theory
• HARK is intended to grow to be for a somewhat more general audience, including programmers and economics from adjacent fields

In light of that:

• As somebody not trained in consumption theory, I could use a little more background in the "Gentle Introduction" and related onboarding materials -Specifically:
  • If you spelled out CRRA utility as "constant relative risk aversion" the first time it's used, rather than use the acronym, it would be easier to follow
  • It's not immediately clear to this non-specialist why 'risk aversion' is relevant in this case, when all the values are deterministic.

A link to lecture notes or some reliable source about this would go a long way to making this more approachable. I've been picking my way through Wikipedia, but that's not ideal.

[mnwhite]

It's meant for entry into computational methods and research using consumption-saving models, but not (at this time) for teaching the models themselves. The materials (mostly) assume that the reader has been through (roughly) the first year of an economics PhD or MS program.

I thought I did spell out CRRA in the gentle introduction notebook, but it's been edited a lot. We can fix that.

In a model with no risk, being "risk averse" isn't meant to capture risk preferences per se. Mathematically, we need the utility function to be concave so that there is an interior solution to the consumption problem: the agent wants to consume some today and some tomorrow. With linear (risk neutral) utility, the solution to a consumption-saving problem with no risk and no artificial borrowing constraint would be to consume everything in one particular period-- probably the very first period or the very last period (if there is one), but it's possible to structure the parameters so that the agent prefers to consume all of their wealth and (future) income in some particular period. Economically, the parameter rho can also be interpreted as the reciprocal of the intertemporal elasticity of subsitution-- the agent's willingness to reallocate between consumption today and consumption next period. When rho --> 0 and we have linear utility, intertemporal elasticity goes to infinity and the agent will shift all consumption into whatever period of time is "best". This is analogous maximizing a linear function subject to a linear constraint: if a unique solution exists, it is guaranteed to be at a corner.

The other piece of it is that the perfect foresight model is a special case / limiting solution of the more general (and interesting) models with real risk. We want to be able to work with the relationships between the risky models and the simple risk free model.

On Sun, Oct 20, 2019 at 3:24 PM Sebastian Benthall notifications@github.com wrote:

I understand that:

• HARK is currently mainly used to teach a specific field of consumption theory
• HARK is intended to grow to be for a somewhat more general audience, including programmers and economics from adjacent fields

In light of that:

• As somebody not trained in consumption theory, I could use a little more background in the "Gentle Introduction" and related onboarding materials -Specifically:
  • If you spelled out CRRA utility as "constant relative risk aversion" the first time it's used, rather than use the acronym, it would be easier to follow
  • It's not immediately clear to this non-specialist why 'risk aversion' is relevant in this case, when all the values are deterministic.

A link to lecture notes or some reliable source about this would go a long way to making this more approachable. I've been picking my way through Wikipedia, but that's not ideal.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=ADKRAFKBUWJZV7H46GHY6GDQPSV5RA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HTAPQ2A, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADKRAFPUYNTUMW2XWOAN7Y3QPSV5RANCNFSM4JCVWAMA .

[mnwhite]

To clarify the last sentence of the large paragraph: "...maximizing a linear function on R^n_+ subject to a linear constraint..." With n non-negativity constraints and 1 linear constraint, the solution to this problem will be all zeros in all index except one (if it's unique).

On Sun, Oct 20, 2019 at 5:24 PM Matthew White mnwhite@gmail.com wrote:

It's meant for entry into computational methods and research using consumption-saving models, but not (at this time) for teaching the models themselves. The materials (mostly) assume that the reader has been through (roughly) the first year of an economics PhD or MS program.

I thought I did spell out CRRA in the gentle introduction notebook, but it's been edited a lot. We can fix that.

In a model with no risk, being "risk averse" isn't meant to capture risk preferences per se. Mathematically, we need the utility function to be concave so that there is an interior solution to the consumption problem: the agent wants to consume some today and some tomorrow. With linear (risk neutral) utility, the solution to a consumption-saving problem with no risk and no artificial borrowing constraint would be to consume everything in one particular period-- probably the very first period or the very last period (if there is one), but it's possible to structure the parameters so that the agent prefers to consume all of their wealth and (future) income in some particular period. Economically, the parameter rho can also be interpreted as the reciprocal of the intertemporal elasticity of subsitution-- the agent's willingness to reallocate between consumption today and consumption next period. When rho --> 0 and we have linear utility, intertemporal elasticity goes to infinity and the agent will shift all consumption into whatever period of time is "best". This is analogous maximizing a linear function subject to a linear constraint: if a unique solution exists, it is guaranteed to be at a corner.

The other piece of it is that the perfect foresight model is a special case / limiting solution of the more general (and interesting) models with real risk. We want to be able to work with the relationships between the risky models and the simple risk free model.

On Sun, Oct 20, 2019 at 3:24 PM Sebastian Benthall < notifications@github.com> wrote:

I understand that:

• HARK is currently mainly used to teach a specific field of consumption theory
• HARK is intended to grow to be for a somewhat more general audience, including programmers and economics from adjacent fields

In light of that:

• As somebody not trained in consumption theory, I could use a little more background in the "Gentle Introduction" and related onboarding materials -Specifically:
  • If you spelled out CRRA utility as "constant relative risk aversion" the first time it's used, rather than use the acronym, it would be easier to follow
  • It's not immediately clear to this non-specialist why 'risk aversion' is relevant in this case, when all the values are deterministic.

A link to lecture notes or some reliable source about this would go a long way to making this more approachable. I've been picking my way through Wikipedia, but that's not ideal.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=ADKRAFKBUWJZV7H46GHY6GDQPSV5RA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4HTAPQ2A, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADKRAFPUYNTUMW2XWOAN7Y3QPSV5RANCNFSM4JCVWAMA .

[sbenthall]

Thank you very much for all this explanation.

It's very helpful (to me) to know that concavity is there to make it better to consume some today, then some tomorrow. That makes sense. Spelling that out, in terms of concavity (as opposed to risk aversion, though I realize these mathematically amount to the same thing), might make the introduction even gentler.

This is a fascinating project to me in part because it gets to the heart of a criticism of formal economics: that it sometimes favors analytical tractability rather than realism in its models. Fixing the analytical tractability problem with software is a great solution to that. But maybe that should lead to a corresponding higher standard in how well motivated the models are

[mnwhite]

To be clear: this project is not meant to teach heterogeneous agents macroeconomics (or structural micro) in itself. HARK is not the beginning of HA macro, nor do these models represent a new frontier of economics. There's a lot of foundational work on models of consumption-saving under risk from the late 80's to mid 90's, a good chunk of it by Chris Carroll, the PI of this project. We aren't trying to provide an entry point into theory, nor justify why these models are a "good" approximation of an individual or household's problem (or at least a better one than prior models). The intent of HARK is to provide a common software package for researchers to work in, rather than everyone handcrafting their own bespoke code for each project. We want to foster interoperability of code between economists; we're starting from nearly zero on that front.

The second goal of HARK (and Econ-ARK more broadly) is to provide teaching tools for computational methods for solving these models. Another project, QuantEcon, provides extensive (and excellent) resources for learning the "building blocks" of computational economics-- numeric methods that are applicable in a wide variety of fields-- and then begins to apply them to some models. Our instructional notebooks are extremely sparse right now, but our intent on that front is to be the next step beyond the QuantEcon material.

I forgot to write in my prior post: Chris' (outstanding) lecture notes are linked several times from Econ-ARK materials; things have been edited a lot, so I don't know off the top of my head where those links are.

On Mon, Oct 21, 2019 at 7:35 AM Sebastian Benthall notifications@github.com wrote:

Thank you very much for all this explanation.

It's very helpful (to me) to know that concavity is there to make it better to consume some today, then some tomorrow. That makes sense. Spelling that out, in terms of concavity (as opposed to risk aversion, though I realize these mathematically amount to the same thing), might make the introduction even gentler.

This is a fascinating project to me in part because it gets to the heart of a criticism of formal economics: that it sometimes favors analytical tractability rather than realism in its models. Fixing the analytical tractability problem with software is a great solution to that. But maybe that should lead to a corresponding higher standard in how well motivated the models are

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=ADKRAFMTBMQOHGHCAOLRTS3QPWHWZA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB2AJSI#issuecomment-544474313, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADKRAFNRWJKTQJOAFTPYZK3QPWHWZANCNFSM4JCVWAMA .

[sbenthall]

Thanks so much for this clarification and introduction to the field.

I've been tasked just now (in the roadmapping meeting) with working on synthesizing/reconciling the introductory materials. I'll keep all this in mind as I do that.

This is one of the documents that made me think that HARK might have a broader aspirational scope, accessible to a somewhat broader audience than the one you indicate, and capable of facilitating more more ambitious intellectual or educational agenda: https://github.com/econ-ark/PARK/blob/master/Elevator-Spiels.md

[sbenthall]

I'm editing the introductory materials (see #409 ) This task is taking me back to this issue.

I think a source of confusion is that there are many imagined audiences/potential users of HARK. I.e., the pool of potential community members is heterogenous.

That imagined heterogeneity is displayed here: https://hark.readthedocs.io/en/latest/readme.html#learning-hark

It looks like the audiences for the "Gentle Introduction to HARK" tutorial is:

• "For people with a technical/scientific/computing background but little economics background"
• "For economists who have not yet done any structural modeling but might be persuadable to start"

So I think I'm going to go with my instincts here and plan to add more background in structural modeling and basic economic theory in the "Gentle Introduction". I hope that's acceptable to you.

It looks like other notebooks are recommended for those with more background in this area.

[llorracc]

Two points:

1. I’m generically sympathetic to criticisms involving too much jargon.
   • The key trick is to find a way to express your point in plain language in a way where the expert will know precisely what you are signaling (basically, they will substitute the jargon term for your plain language term), and the nonexpert will know at least vaguely what you are saying. But this is easier said than done, which is part of the reason jargon is so pervasive.
2. I’d distinguish between jargon (which is about language) and content. You are quite convincing on the point that people who come to the toolkit from outside of economics will find the “Gentle Intro” material an imperfect intro because it leaves out a bunch of epistemological foundations that are assumed (like the importance of expectations). If there are two dimensions (high/low) in familiarity with economics, and high/low in computer stuff, the “Gentle Intro” is kind of written for the high/low (=Econ/Computation) dimension, and you are coming to it from the low/high corner. Seems to me that we are better off just being more explicit about providing different starting points based on prior knowledge. So, we should probably have a (low/high) and a (high/high) to complement the (high/low) intro we have now. I think three paths into the toolkit are enough. (The number of low/low people who will be exploring our toolis is probably — well — low.)

On Tue, Oct 22, 2019 at 8:47 AM Sebastian Benthall notifications@github.com wrote:

I'm editing the introductory materials (see #409 https://github.com/econ-ark/HARK/issues/409 ) This task is taking me back to this issue.

I think a source of confusion is that there are many imagined audiences/potential users of HARK. I.e., the pool of potential community members is heterogenous.

That imagined heterogeneity is displayed here: https://hark.readthedocs.io/en/latest/readme.html#learning-hark

It looks like the audiences for the "Gentle Introduction to HARK" tutorial is:

• "For people with a technical/scientific/computing background but little economics background"
• "For economists who have not yet done any structural modeling but might be persuadable to start"

So I think I'm going to go with my instincts here and plan to add more background in structural modeling and basic economic theory in the "Gentle Introduction". I hope that's acceptable to you.

It looks like other notebooks are recommended https://hark.readthedocs.io/en/latest/readme.html#for-economists-who-have-done-some-structural-modeling for those with more background in this area.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=AAKCK762REU2YSTKKTKVJ5LQP3Y7TA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB5TIKY#issuecomment-544945195, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAKCK73YQNX5SXEVMMNMUY3QP3Y7TANCNFSM4JCVWAMA .

--

• Chris Carroll

[sbenthall]

See #380

[mnwhite]

Agree with Chris here. Multiple entry points might be a better solution, and we need to be clear about which door is for which person. Given the audiences we've presented to, we've often been told that we don't need to explain the economics, just show how it works.

HARK can be used to teach graduate-level dynamic modeling for the first time, but it wasn't our intent.

On Wed, Oct 23, 2019 at 11:21 PM Christopher Llorracc Carroll < notifications@github.com> wrote:

Two points:

1. I’m generically sympathetic to criticisms involving too much jargon.
   • The key trick is to find a way to express your point in plain language in a way where the expert will know precisely what you are signaling (basically, they will substitute the jargon term for your plain language term), and the nonexpert will know at least vaguely what you are saying. But this is easier said than done, which is part of the reason jargon is so pervasive.
2. I’d distinguish between jargon (which is about language) and content. You are quite convincing on the point that people who come to the toolkit from outside of economics will find the “Gentle Intro” material an imperfect intro because it leaves out a bunch of epistemological foundations that are assumed (like the importance of expectations). If there are two dimensions (high/low) in familiarity with economics, and high/low in computer stuff, the “Gentle Intro” is kind of written for the high/low (=Econ/Computation) dimension, and you are coming to it from the low/high corner. Seems to me that we are better off just being more explicit about providing different starting points based on prior knowledge. So, we should probably have a (low/high) and a (high/high) to complement the (high/low) intro we have now. I think three paths into the toolkit are enough. (The number of low/low people who will be exploring our toolis is probably — well — low.)

On Tue, Oct 22, 2019 at 8:47 AM Sebastian Benthall < notifications@github.com> wrote:

I'm editing the introductory materials (see #409 https://github.com/econ-ark/HARK/issues/409 ) This task is taking me back to this issue.

I think a source of confusion is that there are many imagined audiences/potential users of HARK. I.e., the pool of potential community members is heterogenous.

That imagined heterogeneity is displayed here: https://hark.readthedocs.io/en/latest/readme.html#learning-hark

It looks like the audiences for the "Gentle Introduction to HARK" tutorial is:

• "For people with a technical/scientific/computing background but little economics background"
• "For economists who have not yet done any structural modeling but might be persuadable to start"

So I think I'm going to go with my instincts here and plan to add more background in structural modeling and basic economic theory in the "Gentle Introduction". I hope that's acceptable to you.

It looks like other notebooks are recommended < https://hark.readthedocs.io/en/latest/readme.html#for-economists-who-have-done-some-structural-modeling

for those with more background in this area.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub < https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=AAKCK762REU2YSTKKTKVJ5LQP3Y7TA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB5TIKY#issuecomment-544945195 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AAKCK73YQNX5SXEVMMNMUY3QP3Y7TANCNFSM4JCVWAMA

.

--

• Chris Carroll

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=ADKRAFISTZYT7FW3VD57NFLQQEIERA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECDRVAI#issuecomment-545725057, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADKRAFOTQIENTRNUJDPJKO3QQEIERANCNFSM4JCVWAMA .

[mnwhite]

Clarification: It wasn't our intent when making things like the Gentle Intro

On Thu, Dec 5, 2019 at 10:19 AM Matthew White mnwhite@gmail.com wrote:

Agree with Chris here. Multiple entry points might be a better solution, and we need to be clear about which door is for which person. Given the audiences we've presented to, we've often been told that we don't need to explain the economics, just show how it works.

HARK can be used to teach graduate-level dynamic modeling for the first time, but it wasn't our intent.

On Wed, Oct 23, 2019 at 11:21 PM Christopher Llorracc Carroll < notifications@github.com> wrote:

Two points:

1. I’m generically sympathetic to criticisms involving too much jargon.
   • The key trick is to find a way to express your point in plain language in a way where the expert will know precisely what you are signaling (basically, they will substitute the jargon term for your plain language term), and the nonexpert will know at least vaguely what you are saying. But this is easier said than done, which is part of the reason jargon is so pervasive.
2. I’d distinguish between jargon (which is about language) and content. You are quite convincing on the point that people who come to the toolkit from outside of economics will find the “Gentle Intro” material an imperfect intro because it leaves out a bunch of epistemological foundations that are assumed (like the importance of expectations). If there are two dimensions (high/low) in familiarity with economics, and high/low in computer stuff, the “Gentle Intro” is kind of written for the high/low (=Econ/Computation) dimension, and you are coming to it from the low/high corner. Seems to me that we are better off just being more explicit about providing different starting points based on prior knowledge. So, we should probably have a (low/high) and a (high/high) to complement the (high/low) intro we have now. I think three paths into the toolkit are enough. (The number of low/low people who will be exploring our toolis is probably — well — low.)

On Tue, Oct 22, 2019 at 8:47 AM Sebastian Benthall < notifications@github.com> wrote:

I'm editing the introductory materials (see #409 https://github.com/econ-ark/HARK/issues/409 ) This task is taking me back to this issue.

I think a source of confusion is that there are many imagined audiences/potential users of HARK. I.e., the pool of potential community members is heterogenous.

That imagined heterogeneity is displayed here: https://hark.readthedocs.io/en/latest/readme.html#learning-hark

It looks like the audiences for the "Gentle Introduction to HARK" tutorial is:

• "For people with a technical/scientific/computing background but little economics background"
• "For economists who have not yet done any structural modeling but might be persuadable to start"

So I think I'm going to go with my instincts here and plan to add more background in structural modeling and basic economic theory in the "Gentle Introduction". I hope that's acceptable to you.

It looks like other notebooks are recommended < https://hark.readthedocs.io/en/latest/readme.html#for-economists-who-have-done-some-structural-modeling

for those with more background in this area.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub < https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=AAKCK762REU2YSTKKTKVJ5LQP3Y7TA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOEB5TIKY#issuecomment-544945195 , or unsubscribe < https://github.com/notifications/unsubscribe-auth/AAKCK73YQNX5SXEVMMNMUY3QP3Y7TANCNFSM4JCVWAMA

.

--

• Chris Carroll

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/econ-ark/HARK/issues/407?email_source=notifications&email_token=ADKRAFISTZYT7FW3VD57NFLQQEIERA5CNFSM4JCVWAMKYY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGOECDRVAI#issuecomment-545725057, or unsubscribe https://github.com/notifications/unsubscribe-auth/ADKRAFOTQIENTRNUJDPJKO3QQEIERANCNFSM4JCVWAMA .

[sbenthall]

Ok, I see now where to make some impact on this.

I should complete the "Journey into HARK" for engineers. That makes this issue redundant with #362, which I've now claimed. Thanks for your patience with my stumbling around, getting my bearings, and so on.